Correct model for outcome variable with 4 categories

[TKM6403]

I am working on a study that measures the association between BMI(exposure) and multi level(outcome). I am testing with BMI both continuously and categorically, but the outcome is a score with 4 levels. I am having trouble deciding on correct regression analysis for an outcome of 4 levels. If necessary, I can change to dichotomous, but I feel my findings would lose some power. Any suggestions?

 

[GretaGarbo]

Yes, you will loose something if you change the variable to a dichotomous variable. Search and read what Frank Harrel has said about that.

 

[TKM6403]

I wanted to use ordinal logistic regression, but the proportional odds assumption failed. Is there anything else anyone has in mind?

 

[Editor]

You can try with partial proportional odds model